Study protocol: The Maternal Health Multilevel Intervention for Racial Equity (Maternal Health MIRACLE) Project

Abstract

Purpose.

To test the effectiveness and cost-effectiveness of a multilevel intervention for population-level African American (AA) severe maternal morbidity and mortality.

Background.

Severe maternal morbidity and mortality in the U.S. disproportionately affect AA women. Inequities occur at many levels, including community, provider, and health system levels.

Design.

Intervendtion.

Throughout the two intervention counties, we will expand access to enhanced prenatal care services using telehealth and flexible scheduling (community level), provide actionable maternal health-focused anti-racism training (provider level), and implement equity-focused community care maternal safety bundles (health system level).

Partnership.

Interventions were developed/co-developed by intervention county partners, including AA women, enhanced prenatal care staff, and health providers. For equity, 46% of project direct cost dollars go to our partners. Most study investigators are female (75%) and/or AA (38%). Partners are overwhelmingly AA women.

Sample, measures, analyses.

We use a quasi-experimental difference-in-differences with propensity scores approach to compare pre (2016–2019) to post (2022–2025) changes in outcomes for Medicaid-insured women in intervention counties to similar women in the other Michigan, USA, counties. The sample includes all Medicaid-insured deliveries in Michigan during these years (n~540,000), with women observed during pregnancy, at birth, and up to 1 year postpartum. Measures are taken from a linked dataset that includes Medicaid claims and vital records.

Conclusion.

This study is among the first to examine effects of any multilevel intervention on AA severe maternal morbidity and mortality. It features a rigorous quasi-experimental design, multilevel multi-partner county-wide interventions developed by community partners, and assessment of intervention effects using population-level data.

1. Introduction

The U.S. maternal mortality rate is the highest among high-income countries.^1,2 Pregnancy-related mortality rates doubled between 1987–2016 in the U.S., while declining in other high-income countries.^1,2 Around 60% of the deaths are preventable.³ Of all deaths, approximately 1/3 occur during pregnancy, 1/3 at delivery or the week after, and 1/3 up to 1 year postpartum.²

Severe maternal morbidity (SMM; “unexpected outcomes of labor and delivery that result in significant short- or long-term consequences to a woman’s health”⁴) affects another 60,000 US women every year.⁵ The most common SMM include procedures in which women receive blood transfusions around delivery, hysterectomy, and ventilation/tracheostomy.² These life-threatening complications affect mothers, families, and communities, and cost billions of dollars per year.⁶

SMM and maternal mortality disproportionately affect African American (AA) women. AA women are three to four times more likely to die of pregnancy related complications and have twice the SMM rate compared to non-Hispanic white (NHW) women.^7,8 Inequities affecting SMM occur at multiple levels¹⁰ (e.g., community, health system, and provider/practice levels), yet very few studies examine multilevel interventions for SMM and maternal mortality disparities. The U.S. National Institutes of Health has called for additional research on multilevel interventions to reduce SMM and maternal mortality disparities.⁹

This protocol paper describes a study testing effectiveness and cost-effectiveness of a multilevel intervention (i.e., intervening at the community level, provider/practice level, and health system level) to reduce AA SMM and maternal mortality in two Michigan, U.S. counties. The Maternal Health Multilevel Intervention for Racial Equity (Maternal Health MIRACLE) Project uses a quasi-experimental difference-in-differences (DiD) with propensity scores approach to compare pre (2016–2019) to post (2022–2025) changes in outcomes among Medicaid-insured women in the two intervention counties with similar women in other Michigan counties. The study sample will include all Medicaid-insured deliveries in Michigan from 2016–2019 and 2022–2025 (~540,000 births, including ~162,000 births to AA women). Specific aims are to:

1. Assess the effectiveness of the multilevel intervention on:

   1. AA SMM and a composite of SMM and pregnancy-related mortality (overall & relative to NHW women)
   2. AA non-severe maternal morbidity (overall & relative to NHW women)

2. Test improved service utilization and non-severe maternal morbidity (overall & relative to NHW women) as mechanisms (i.e., mediators) of the intervention’s effect on SMM

3. Evaluate the cost-effectiveness of the intervention

This study is among the first to examine effects of any multilevel intervention on AA SMM and mortality, and is the first to test this specific multilevel intervention. Novel aspects of the study include its rigorous quasi-experimental design, use of interventions at each level that were developed or co-developed by community partners in experimental counties, assessment of intervention effects using population-level data, and use of a unique statewide database system to assess study outcomes.

2. Method

2.1. Non-randomized pragmatic trial.

A non-randomized pragmatic trial design was chosen because randomizing individuals within counties to receive or not receive services associated with state or federally funded programs (such as an enhanced prenatal and postnatal care telehealth option) is not allowed by federal rules. Furthermore, individual-level randomization does not make sense for county-level interventions. County-level randomization is not possible because our intervention counties developed the interventions and thus would have to be in the intervention condition. The use of DiD with period-specific propensity score kernel weighting allows for effectiveness analyses using a non-randomized design, relying on observed individual and neighborhood characteristics to minimize bias.

2.2. Conceptual model and pilot work.

Disparities in maternal outcomes persist after controlling for patient characteristics (e.g., preconception health, health behaviors) and health care system factors (e.g., site and quality of care), suggesting that additional factors may be contributing to the high prevalence of maternal morbidity.⁹ The goal of our multilevel intervention is to help perinatal health systems engage and serve AA women more effectively to reduce disparities. The conceptual model framing our intervention merges the multiple levels where inequities occur highlighted by Howell¹⁰ and the right to health framework used by Black Mamas Matter Alliance^11,12 (Figure 1). As described by Howell,¹⁰ inequities occur at multiple levels.

Figure 1. Conceptual model merging Howell multilevel inequities¹⁰ and the Right to Health Framework^11,12.

For explanation of abbreviations, see Table 1.

2.2.1. Community level.

AA women disproportionately face community and neighborhood barriers to care.^10,13–15 Working multiple jobs, limited access to quality childcare, unsafe neighborhoods, and transportation barriers can make it difficult to visit an office or accept home visiting services such as enhanced prenatal and postnatal care (EPC). EPC programs (such as Healthy Start and Michigan’s Maternal Infant Health Program) improve maternal care¹⁶ and reduce adverse birth outcomes, especially for AA women.^17–21 However, up to two-thirds of eligible women nationally do not participate in EPC programs.^22,23 In our pilot work before the COVID-19 pandemic, half of 50 interviewed women who declined these services said they would reconsider if they were offered via telehealth and with more flexible scheduling, including evenings and weekends. By offering EPC services via telehealth at flexible hours to those who decline traditional in-home services, we anticipate restructuring these services in a way that better fits the lived realities of many AA women. The goal is to modify community-based services to better accommodate the logistical challenges that women in low-income communities face.

2.2.2. Provider/practice level.

Clinicians’ explicit and implicit racial beliefs and attitudes are linked to quality of care,^24,25 suggesting that racial bias in the health care system contributes to disparities in maternal morbidity and mortality.⁹ When AA women experience racial discrimination and inadequate care, especially during vulnerable moments in their health,^26,27 this leads to adverse outcomes, mistrust of providers, and decreases likelihood of engaging health services in the future.^28,29 We will focus on changing provider and health system biases and corresponding structures/practices using an actionable, maternal health focused anti-racism training. This training was developed and tested by AA women in Genesee County and found to be feasible, acceptable, and to result in improvements in provider knowledge, understanding of the effects of racism on maternal health, and familiarity with structural challenges AA women face and specific actions they can take to improve AA maternal health.^14,30,31

2.2.3. Health system level.

Overall improvements in quality of obstetric care do not necessarily translate to reductions in racial disparities in maternal outcomes.^9,32 Therefore, it is important for maternal care quality improvement interventions to target racial disparities.⁹ The National Healthy Start Association, with support from the U.S. Health Resources and Services Administration, has launched the Alliance for Innovation on Maternal health Community Care Initiative (AIM-CCI) to develop and implement equity-focused maternal safety bundles for community-based organizations and outpatient clinical settings.³³ Our health system-level intervention will implement these equity-focused maternal care quality improvement bundles in intervention counties.

2.2.4. Proposed mechanisms.

We propose that (a) improved service utilization, and (b) reduced non-severe maternal morbidity will serve as mechanisms of the effect of the intervention on AA SMM. We use y/n enrollment in EPC, more (prenatal and postnatal) outpatient visits, and fewer (prenatal and postnatal) emergency department (ED) visits as positive service utilization indicators (Figure 1). NSMM will be defined as the sum of WHO-delineated NSMM diagnoses and procedures (details in Section 2.10.3). Number of NSMM will be tested as a mediator of the effects of study interventions on SMM because reducing NSMM (e.g., pre-eclampsia) reduces risk of SMM (e.g., eclampsia).

2.3. Equity and community engagement.

This project engages three kinds of partners. AA women from the target counties developed, implemented, honed, tested, and will lead the provider/practice level actionable antiracism intervention.^{14,30,34–37} EPC staff and leaders co-developed and will lead the community-level telehealth intervention. They are one of six national pilot sites for the AIM-CCI community care maternal safety bundles (our system-level intervention). Providers/health systems have provided input and/or feedback on interventions at all three levels. Representatives from all of these communities are valued members of the study team.

We have attended to study equity and representation. In this project, 46% of the direct cost dollars go to our partners. 75% of the study investigators are women; 38% of study investigators are AA. Partners are overwhelmingly AA and female.

2.4. Comparison condition: usual care.

Pregnant AA women in comparison counties (all Michigan counties except the two intervention counties) will receive whatever EPC services (i.e., the Maternal Infant Health Program and/or Healthy Start) they naturalistically choose to receive.

2.4.1. Maternal Infant Health Program (MIHP) as usual.

All pregnant women in Michigan who are Medicaid insured are eligible for MIHP. MIHP is an EPC program that offers at least monthly home visiting and care coordination to enhance medical care during pregnancy and up to 12 months post birth. MIHP offers care coordination by nurses and social workers, risk assessment, individual care plans, interventions matched to risk factors, and referrals (e.g. to address depression, domestic violence, health behaviors, basic needs, contraception). Before the COVID-19 pandemic, women were primarily served in their homes, with telehealth being the only option early in the pandemic. In-home visits are, again, becoming the norm. MIHP improves maternal and infant care,¹⁶ reduces risk of adverse birth outcomes,¹⁷ and reduces infant mortality.³⁸ However, statewide, less than one third of women who are eligible enroll in MIHP.³⁹

2.4.2. Healthy Start as usual.

Healthy Start is a federally funded program for women of color. It offers more intensive EPC services delivered by race/ethnicity concordant community health workers, whoare expected to have face-to-face contact every other week in women’s homes. Community health workers offer peer support; resilience and problem solving; risk assessment; facilitating provider-client communication; collaborative care; system navigation, including transition from prenatal care to postnatal primary care; and supportive referrals.²⁷ Healthy Start is effective in reducing adverse birth outcomes. ^1018,19 Healthy Start is offered in six of Michigan’s highest risk communities for AA infant mortality, including the two intervention and four of the comparison communities.

2.5. Intervention.

In addition to traditional EPC services, the following county-wide manualized intervention components will be offered in both intervention counties (see Table 2). The three intervention levels operate independently of each other.

Table 2.

Study interventions and assessments (whole table is new)

County-wide intervention component and years offered	Offered in control counties?	Offered in intervention counties?	Intervention fidelity assessment data collected by the project team (frequency)	Outcome assessment data from state records (frequency)
EPC (MIHP and Healthy Start) services as usual (all years)	Yes	Yes	NA	All linked Medicaid claims and vital records data from the entire state of Michigan for pre (2016–2019) and intervention (2022–2025) years
Community level: EPC services also available via telehealth and outside business hours for women who decline EPC services as usual (2022–2025)	No	Yes	• Number of AA and non-AA women who declined traditional EPC services, who agreed to telehealth services, and who were served (monthly)
Provider/practice level: Maternal health focused actionable anti-racism for health-related providers and staff (2022 – 2025)	No	Yes	• Number trained (each training) • Provider pre, post, and 1 month follow-up surveys (each training) • Penetration (# trained/# eligible providers in county; end of study)
Health system level: Implementation of AIM-CCI’s equity focused community care maternal safety bundles (2022 – 2025)	No	Yes	• Case notes documenting implementation activities (continuous) • Survey of knowledge and use of AIM-CCI bundle topics for the year (i.e., reach and adoption; annual)

2.5.1. Community level (improving accessibility).

We will make EPC services (i.e., MIHP and Healthy Start) available via telehealth with flexible hours to women who are eligible for Healthy Start (primarily women of color) but who decline traditional MIHP or Healthy Start services. We do not anticipate that this intervention will change the dose or activities of EPC services, but rather that it will allow women to engage in EPC services who would otherwise decline. “Telehealth” EPC will include telehealth initially but could include face-to-face contact if desired by the participant. The goal is to have inclusive options to address patient preferences. We met with EPC providers to adapt core program components to telehealth delivery (e.g., mailing plans of care, demonstrating breastfeeding during video visits).

2.5.2. Provider/practice level (improving acceptability).

Our goal is to provide actionable maternal health-focused anti-racism training to anyone in intervention counties interacting with AA perinatal women related to their health, including health system administrators, physicians, residents, midwives, nurses, front desk staff, schedulers, public health officers, EPC staff, doulas, WIC staff, and lactation consultants (n~480).

The daylong actionable maternal health focused anti-racism training, occurring in groups of 10–20 every 6 weeks, will be led by the AA community partners who previously developed and led this training.^14,31 Training includes a virtual tour of local neighborhoods to illustrate structural assets and challenges. Training will use actionable anti-racism and windshield tour manuals,^14,31 which include discussion, reflection, and experiential activities covering: (1) health disparities and social determinants of health through an AA historical/cultural lens; (2) African and AA history that created current neighborhood and social structures; (3) historical, cultural, and structural impacts of racism and implications for maternal health and maternal mortality; (4) structural challenges and assets in high AA neighborhoods in each county and how they affect health; (5) how participating providers’ health and public health practices can better “meet women where they’re at,” and be respectful, flexible, and more responsive to their needs; (7) how to engage perinatal AA women as partners in improving health systems that serve them; and (8) setting actionable goals for their settings. Modules help participants brainstorm ways to tailor their practices to be more responsive to perinatal AA women.

2.5.3. Health system level (improving quality).

Kent County is one of six communities nationally (and the only one in Michigan) piloting AIM-CCI’s equity focused community care maternal safety bundles as they are developed by the national AIM-CCI group each year. Community care is defined as care provided by outpatient, EPC, community-based organizations, and linkages between hospital care and these settings. As part of AIM-CCI, Kent County planned to implement 2 bundles per year for 4 years (8 total) within outpatient clinical settings and community-based organizations. As part of this project, Genesee County will implement the bundles in partnership with Kent. Examples of planned bundles include “transition from maternity care to well-woman care,” “comorbidities,” and “maternal mental health.” The first bundle to be implemented was called “postpartum care basics for maternal safety.” It consisted of county-wide efforts to: (1) create pathways for systematic referral of all Medicaid-eligible pregnant women to EPC services; and (2) use the Center for Disease Control and Prevention’s (CDC’s) HEAR HER program,⁷⁶ which identifies NSMM and SMM warning signs and appropriate responses for providers, patients, staff, and family members.

The charge from the national AIM-CCI initiative to the pilot sites is to “implement” the bundles. Our study team will partner with the county implementation teams to bring implementation science expertise to the process. We will manualize and systematize local implementation approaches, using Replicating Effective Programs⁴⁰ to guide implementation activities and the RE-AIM framework⁴¹ to guide implementation assessment. We will also document implementation processes (see next section).

2.6. Intervention fidelity assessment.

At the community (telehealth option/flexible hours) level, we will ask our intervention county EPC (i.e., MIHP/Healthy Start) partners to report the following to us monthly: number of AA and non-AA women who declined traditional EPC services, who agreed to telehealth services, and who were served with these services.

At the provider/practice (actionable racism training) level, we will track: the number of health care providers participating in the trainings, penetration of these activities (number of providers trained divided by number eligible in each county),⁴¹ and electronic pre, post, and 1 month assessments of provider knowledge (e.g., of AA history, how race and racism can affect health) and expected/actual actions (e.g., I have changed something about my practice to reduce AA SMM and maternal mortality).^14,30,31,42

At the system (AIM community care maternal safety bundle implementation) level, we will document implementation activities undertaken in both intervention counties using a structured implementation case note.^43–47 Notes will document the actors, recipients, purpose, encounter length, a checklist of implementation strategies used (from Powell et al., 2015),⁴⁸ a checklist of discussion topics, and free response sections. We will conduct annual county-wide electronic surveys asking about recipients’ knowledge and use of the AIM-CCI bundle topics for that year (i.e., bundle reach⁴¹) and actions taken to address each bundle guideline (i.e., bundle adoption⁴¹).

2.7. Study sample.

The study sample (n~540,000) will consist of all Medicaid insured deliveries in Michigan in 2016–2019 (pre-intervention cohorts) and 2022–2025 (post-intervention cohorts; 2020–2021 are transition years excluded from analyses). Medicaid covers 43% of all and 66% of AA births in the U.S.,^49,50 disproportionately women who experience SMM or maternal mortality.^51–53 Maternal race will be assessed based on birth records. County of residence at delivery (in data routinely collected by the state of Michigan; details below) will be used to assign women to intervention (i.e., Genesee and Kent counties) and comparison (i.e., all other Michigan) counties. Intervention counties and the rationale for choosing them are described in Table 3.

Table 3.

Description of Intervention Counties^*

	Genesee County	Kent County	Michigan
Population	406k	657k	9,987k
Births per year (all)	4.5k	8.6k	110k
% Medicaid (all)	56%	31%	42%
Births to AA women	1.4k	1.3k	22k
% Medicaid (AA)	83%	61%	66%
Main metro area	Flint	Grand Rapids	--
% AA in metro	55%	20%	14%
Poverty in metro	40%	21%	14%

^*We chose Genesee and Kent counties as intervention counties because: (1) all levels of the multilevel intervention were developed in these counties; (2) both have Healthy Start; (3) both counties have high maternal morbidity and infant mortality;⁷⁵ and (4) the counties represent different areas of Michigan (i.e., Eastern vs. Western Michigan).

2.8. Sample size and power.

The sample will include over 60,000 Medicaid insured deliveries in the intervention counties and close to 480,000 in the comparison counties over the 8 pre- and post-intervention birth cohorts. Approximately 30% of the deliveries will be AA women and 60% NHW. The SMM prevalence (main outcome) among Medicaid-insured deliveries in 2010 was 166/10,000;⁷ we assumed conservatively that it remained constant. We assumed for power calculations that the AA-NHW SMM prevalence ratio is 2 to 1. This study is powered to detect DiD intervention effects on reducing SMM among deliveries to AA women as small as relative risk of 0.84 with 80% power and 0.05 statistical significance assuming an intra-cluster correlation coefficient of 0.2. Because maternal mortality, even at a state level, is rare, our secondary outcome of maternal mortality will be accounted for as a composite of (SMM + maternal mortality; 1 if either, 0 otherwise).^54,55 The available sample will allow us to detect smaller effects in disparities and in secondary outcomes, which include binary and non-binary assessments.

2.9. Data sources.

We will rely on the Michigan Department of Health and Human Services (MDHHS) Health Services Data Warehouse. Linked sources of data, all routinely collected or publicly available (e.g. census tract data), include: (1) complete Medicaid pregnancy and postpartum (12 months after birth) medical claims, including International Classification of Diseases (ICD-10) diagnostics and procedures; (2) monthly Medicaid enrollment during pregnancy and postpartum; (3) birth records, including data on pre-pregnancy and pregnancy risk factors (e.g. chronic disease, prior preterm); (4) maternal death records; (5) additional program data, including income level and participation in other programs (e.g. cash assistance); and (6) census tract data.

2.10. Assessments.

2.10.1. Covariates for propensity score estimation

will include age, education, married(y/n), father identified, alcohol use, tobacco use, prior preterm birth, a previous birth within 18 months of conception, and WIC participation, all from birth records via the MDHHS Health Services Data Warehouse. Medicaid eligibility and claims will be used to create an indicator for having Medicaid coverage three months prior to pregnancy. Census variables at the block group and census tract level will be used to adjust for poverty and household characteristics (e.g. % families in poverty, % families that are AA, Childhood Opportunity Index⁵⁶, Area Deprivation Index^57,58). An indicator will identify women living in counties where Healthy Start is offered.

2.10.2. Primary outcomes.

We will assess severe maternal morbidity (SMM, as defined by the American College of Obstetricians and Gynecologists and the CDC)⁴ using CDC’s list of 21 SMM indicators based on ICD-10 diagnosis and procedure codes.⁵⁹ The binary overall SMM indicator will be coded 1 if any SMM is identified in antepartum, intrapartum, or 12 months postpartum Medicaid claims and 0 otherwise. Maternal death will be defined using the standard CDC definition of death from a pregnancy-related cause within one year of delivery or termination of pregnancy,² a binary indicator defined as 1 if maternal death is identified on a death record linked to the pregnancy or delivery, and 0 otherwise. AA-NHW disparities will be defined as the percent difference between AA and NHW outcomes.

2.10.3. Secondary outcomes.

Measurement of “non-severe” maternal morbidity (NSMM), is taken from the WHO’s Maternal Morbidity Working Group.^60–63 Most NSMM are measurable using ICD-10 diagnostic and procedure codes^62,64 To capture multimorbidity in our sample, the overall NSMM indicator will be the sum of WHO-delineated NSMM diagnoses and procedures^62,64 identified in a woman’s Medicaid claims during pregnancy and up to 1 year postpartum. We will also assess the following WHO-defined subcategories: (1) number of direct NSMM diagnoses and procedures (e.g., delivery complications, hypertensive disorders of pregnancy, obstetric hemorrhage, pregnancy-related infections);⁶² (2) number of indirect NSMM diagnoses and procedures (e.g. endocrine, nutritional, and metabolic diseases; mental disorders); and (3) number co-incidental NSMM (e.g. partner violence, sexual assault) diagnoses and procedures.⁶²

2.10.4. Proposed mechanisms.

Service utilization will be operationalized using three separate indicators drawn from Medicaid claims and EPC program data, adjusted for number of months of Medicaid coverage: (1) EPC participation (MIHP and/or Healthy Start) defined as a binary indicator;¹⁶ (2) number of outpatient pregnancy and postpartum visits (higher is better); and (3) number of ED visits (lower is better). NSMM will be operationalized as the sum of all NSMM conditions, as described above.

2.10.5. Cost-effectiveness measures.

Our grant accounting will capture the costs of providing the multilevel intervention. The primary outcome used for cost-effectiveness analyses will be SMM. Prevented SMM will be monetized using Medicaid claims data by calculating the difference between Medicaid delivery expenditures between women with SMM and without SMM using our own claims data and prior estimates.⁶ Secondary cost-effectiveness measures will be maternal mortality and NSMM. The value of a statistical life,⁶⁵ currently around $10 million, will be used to monetize prevented maternal deaths. For NSMM, we will calculate intervention costs per point of NSMM score reduction. Costs (and savings) in future years will be discounted to present value in the year of treatment initiation using a 3% discount rate.⁶⁶

2.11. Analyses.

Tests will be two-sided (p<=0.05) and we will report measures of clinical significance (i.e., effect sizes).

2.11.1. Difference-in-differences (DiD) and propensity score methods.

The intervention evaluation will use DiD with period-specific propensity score kernel weighting.^67–69 The standard DiD method ameliorates potential selection bias by subtracting the difference in outcomes between intervention and comparison groups at the pre-intervention, baseline, period (i.e., the systematic difference) from the difference in outcomes between intervention and comparison groups after the implementation (post-intervention). To minimize the potential bias induced by groups systematically changing over time, as the intervention counties are anticipated to increase participation in EPC of high-risk women through telehealth, we will use a propensity score weighted DiD method that balances the groups pre- and post-intervention. The method will ensure that the pre-comparison, post-comparison, pre-intervention, and post-intervention groups will have similar demographic, geographic, and medical backgrounds. Using period-specific propensity score kernel weighting in the DiD analysis is better than standard matching only at baseline because: (1) the latter may create regression to the mean problem⁷⁰, and (2) it does not rely on a limited number of control individuals for each treated woman, allowing a larger sample to be leveraged. We will account for within-patient correlation (i.e., women with repeat pregnancies) using standard statistical techniques.

2.11.2. Missing data.

Data on pregnancy-related mortality is anticipated to be complete. Outcomes defined based on claims up to one year postpartum will be defined per the number of months with Medicaid coverage assessed. All women in Michigan retain Medicaid insurance at least eight weeks after birth, and most (~70%) retain Medicaid coverage through the first postpartum year.

2.11.3. Estimation procedures and effect testing.

The DiD equation is set up as follows, with Y being the outcomes of interest, the intervention indicator taking the value 1 if the observation is in the intervention counties, and 0 if in comparison counties and the post indicator taking the value 1 if the observation is in 2022–2025, and 0 if in 2016–2019. $Y={\beta }_0+{\beta }_1\times intervention+{\beta }_2\times post+{\beta }_3\times intervention\times post+\epsilon$ The key parameter to our analysis is ${\beta }_3$, the DiD effect of the intervention. Propensity score period-specific kernel weights will be used in the DiD regression estimation so that the 4 groups (intervention pre, intervention post, comparison pre, comparison post) will be similar⁷¹ on the set of covariates described above. For each outcome of interest, we will first estimate the intervention effects among AAs. Then, we will estimate the effects on the AA-NHW disparities by adding an interaction term between the AA-NHW race indicator and ${\beta }_3$.

2.11.3.1. Aim 1. Intervention main effects.

We will use the strategy described above to test intervention main effects on: (1) AA SMM (primary); (2) AA SMM and pregnancy-related mortality (1 if either; 0 otherwise); (3) AA-NHW disparities in SMM; (4) AA-NHW disparities in (SMM + mortality); (5) AA overall NSMM; (6) AA direct NSMM; (7) AA indirect NSMM; (8) AA co-incidental NSMM; (9) AA-NHW disparities in overall NSMM; (10) AA-NHW disparities in direct NSMM; (11) AA-NHW disparities in indirect NSMM; and (12) AA-NHW disparities in co-incidental NSMM.

2.11.3.2. Aim 2: Mechanisms of intervention effects.

We will separately test the hypotheses that, relative to AA women from comparison counties, AA women from intervention counties will have better service utilization (three separate measures: EPC participation = yes, more outpatient visits, fewer ED visits) and less NSMM, our proposed mechanisms. We will then test the hypothesis that service utilization and NSMM (1) predict AA SMM, and (2) mediate the effects of the intervention on AA SMM in a structural equation model framework. Statistical significance of the indirect effect will be assessed using bias-corrected bootstrapped standard errors. ⁷²

2.11.3.3. Aim 3: Cost-effectiveness analyses.

We will conduct cost analysis and cost effectiveness analyses of the multilevel intervention relative to the status quo in non-intervention counties. We will conduct the cost effectiveness analyses both from the payer perspective, focusing on costs directly incurred to implement the program, and from the societal perspective, considering all costs regardless of source.

2.11.3.4. Moderators.

We will explore Hispanic ethnicity, age, first pregnancy, education, marital status, and prenatal chronic disease burden (additive score of disease indicators) as moderators of the effects of the intervention on AA SMM.

Conclusion

AA women face addressable disparities in maternal morbidity and mortality. This trial will help build the evidence base to address these disparities. The study will provide cost-effectiveness information needed to drive policy and practice decisions (e.g., sustained Medicaid reimbursement of telehealth EPC, national scale up of AIM-CCI bundles). It will also provide information about mechanisms of intervention effects needed to advance science.

One of the unique factors of this study is its focus during the design, execution, and dissemination of results on inclusion, representation, and community participation. The interventions being tested were designed by or in equal partnership with their end-users (including AA women, EPC staff and leaders, and providers from intervention counties). Interventions are offered in a way that continues this engagement; community partners lead all three interventions with support from the study team. In fact, academic partners wrote the study grant application because they knew of community efforts that were important to their community partners that could use additional effort and financial and technical support. Community partners brought interventions, commitment to finding ways to address pressing practical and clinical issues, and knowledge of community, system, and policy priorities. Academic partners brought a knowledge of the research literature, conceptual models, and experience with research design and grant-writing. During proposal preparation, community partners met twice per week with academic partners, discussed their interests in the project, and agreed to leverage their social capital to help meet study aims. This partnered approach ensures relevance, increases community ownership, and enhances sustainability.⁷⁷ Most population-level studies involving multi-level interventions do not include this level of community engagement or attention to the representativeness of the study team; our study provides an example others can follow.

This project adheres to rigorous standards for causal inference (the quasi-experimental design, use of DiD with propensity score methods, reporting assumptions for creating propensity scores and the resulting balance and overlap).⁷³ Rigor and reproducibility are also ensured by clear research questions, a priori data analysis plans, use of a large linked Medicaid and vital records dataset, reliable and valid measures, transparent statistical and power analyses, sufficient power, valid statistical methods to address missing data, detailed intervention manuals, and fidelity evaluation. Potential limitations include unobserved factors that can induce bias after propensity score matching using observable characteristics, potential incomplete or inaccurate observations in administrative data, and inability to determine effects of each intervention level separately.

It is unusual to try to assess population-level effects of a multilevel intervention mounted by an investigative team. This study provides an example of how to do so.

Table 1.

Glossary of Terms and Abbreviations

AA: African-American

AIM-CCI: Alliance for Innovation on Maternal health Community Care Initiative (AIM-CCI) is an effort by the National Healthy Start Association, supported by the U.S. Health Resources and Services Administration, to develop and implement non-hospital, community care maternal safety bundles. CDC: Centers for Disease Control and Prevention

Community health workers: trained, often race-concordant, peers who are employed to provide health education, support, and advocacy. As trusted members of the community, they facilitate access to services and improve the quality and cultural competence of service delivery. COVID-19: Coronavirus Disease 2019

Difference-in-differences (DiD): quasi-experimental approach to estimate intervention effects by comparing changes over time in intervention vs. control group.

ED: emergency department

Enhanced prenatal and postnatal care (EPC): targets improved birth outcomes through home visiting, care coordination, education, and support. In Michigan, the two primary EPC providers are:  Maternal Infant Health Program (MIHP): home visitation program for Medicaid eligible women and infants in Michigan  Healthy Start is a federal program dedicated to reducing disparities in maternal and infant health in high risk communities

ICD-10: International Statistical Classification of Diseases74

MDHHS Health Services Data Warehouse: administrative linked dataset maintained by MDHHS, including vital records, Medicaid claims, and other data

Michigan Department of Health and Human Services (MDHHS): State of Michigan Department that oversees health policy, population health, and provides public assistance and child/family welfare

MIRACLE: Maternal Health Multilevel Intervention for Racial Equity

NHW: Non-Hispanic white

Non-severe maternal morbidity (NSMM): any health condition attributed to or complicating pregnancy or childbirth that have a negative impact on the woman’s well-being and functioning (e.g., gestational hypertension, obstetric complications, infections, endocrine, nutritional, and metabolic diseases, mental disorders).61–63

Propensity score methods: statistical method to estimate intervention effects in the absence of randomization RE-AIM: reach, effectiveness, adoption, implementation, maintenance

Severe maternal morbidity (SMM): unexpected outcomes of labor and delivery that result in significant short- or long-term consequences to a woman’s health (e.g., eclampsia).